Networks can be used to connect storage devices to computing devices (e.g., clients, servers, and the like). For instance, in a Storage Area Network (SAN), a Fibre Channel network is used to connect computing devices to storage.
In a typical network-based storage environment, all computing devices have access to the available storage devices. Connectivity among the computing devices and the underlying storage devices within the storage environment is shared. This approach provides a wide variety of benefits, including more efficient server platform fail-over. That is, a failed storage device can be replaced automatically by another operational server platform without the need to change cabling to the storage devices.
A new class of distributed computer application has been developed to share access to storage devices across server platforms. These applications seek to use the shared connectivity afforded by SAN technology to share simultaneous access to data at I/O rates that are consistent with the speed of the SAN network. Prior to the development of SAN technology, local and wide area networks provided connectivity between computing devices that did not include storage devices. Connections were established with network protocols such as Transmission Communication Protocol (TCP), Unreliable Datagram Protocol (UDP), and others.
Distributed File Systems such as network file system (NFS) and common Internet file system (CIFS) were layered on top of the network protocols. Distributed File Systems mediate shared access to files across a network. The services provided by distributed file systems are, however, not without significant performance cost. While access to data may be transparent, the rate at which data can be transported between client and server in a distributed file system is limited by the high overhead of managing communication protocols. For instance, the overhead of managing communication protocols limits application I/O rates to a level far below what can be achieved to storage devices that are directly connected to the server platform. Because of this limitation, only applications with relatively low I/O rates can share data using distributed file systems.
SAN systems make storage devices accessible to multiple server platforms and, often, the data stored is accessed by more than one application. One strategy for ensuring the integrity of shared data in a SAN environment is to stabilize (or freeze) a storage object (such as a file system or volume) on one server platform and then to allow access to the same object on another server platform.
Various strategies can be employed to ensure that a disk object remains frozen between two points in time. The simplest method of keeping a disk object frozen is to change the mode of a file system to read-only. This is a drastic and awkward process because the file system is unusable until the remote component completes its work.
Another scheme is distributed lock management. A semaphore is established that can be shared across platforms or among applications. Before mapping, a lock is taken on the object and is retained until relinquished by the remote machine. Distributed lock management has the advantage of arbitrarily fine scale because the semaphore can be designed to encompass individual bytes if necessary. The overhead of managing locks, however, can become cumbersome, and can hinder performance. Locking mechanisms can also block application access to data for long periods and may lead to deadlocks.
The most prevalent strategy for stabilizing disk images is the use of snapshots and mirrors. These mechanisms have the advantage of imposing the least impact on the application because they can be invoked very rapidly. The images created by snapshot and mirror will be referred to collectively as frozen images.
As the storage environment becomes more complex, so does the difficulty of generating a frozen image. A storage environment may consist of many layers of storage objects, or abstractions. For instance, a storage object may be a file system built on top of a volume that is made up of many storage devices. Or a storage object may be distributed across many storage devices, or may consist of file systems built on volumes on a large number of storage devices. The complexity of the storage environment grows dramatically with the number of file systems, volumes and devices, and the choices faced while creating a frozen image within such environments grow proportionately with that complexity. Applications which need to quiesce or to freeze all or portions of the storage environment are faced with the complex job of understanding every aspect of the storage environment.
For the reasons stated above, and for other reasons stated below which will become apparent to those skilled in the art upon reading and understanding the present specification, there is a need in the art for a system and method for controlling the formation of stable images of a diverse group of logical storage objects.